In conversations with different frontier labs over the past couple of weeks, I’ve heard the same refrain: They are worried they might not have spent enough money on compute, either because they didn’t see demand surging as fast as it has, or they just failed to lock up the energy production, chip supplies, or data center leases they now need.

This is a massive shift from late last year, when talk of an AI bubble augured capital expenditures that would come back to bite tech companies when token demand plummeted.

Now it looks like the opposite has happened. (Not that we are all that surprised). Even if companies wanted to spend irresponsible amounts of money on AI compute, they’ve been hampered by supply chain constraints. And the Iran war creates even more friction.

The shortage is actually forcing companies to stay focused. OpenAI shutting down Sora in order to reallocate compute resources is just one example. Other companies are being forced to make trade-offs when deciding where to place their bets.

Constraints can often be good for innovation (necessity is supposed to be the mother of invention after all). When capital is tight, for example, it’s usually a forcing mechanism that rewards only the best companies. With the capital spigot still wide open, the compute bottleneck might be the next best thing for major tech breakthroughs.